Pick the job — three questions
It must be yes to all three. The classic mistake is choosing something too ambitious, watching it go badly, and deciding the whole idea is oversold.
You watched an assistant book a squash court, vet a lead and file a slip. This page is how you start on Monday: pick one job, write it down properly, and run it beside yourself for a fortnight. Everything below came from the talk.
It must be yes to all three. The classic mistake is choosing something too ambitious, watching it go badly, and deciding the whole idea is oversold.
The test of a good brief: could you hand this document to a capable new employee on their first morning and walk away? If yes, it will work. If no, the gap is in how the business explains itself — not in the technology.
One paragraph. What arrives, what should leave, and how you would know it went well.
Two real examples from your own files: one you were happy with, one you sent back — and say why.
What must never happen. Never quote below this. Never commit to a date. Never mention the other client.
The situations where you want a question rather than an attempt. Be generous with these at the start.
The folder, the naming, the system, the person it goes to next.
When it gets something wrong, do not fix the output. Fix this document. That single habit is the difference between a party trick and a system.
The useful question is never “can it do this?” It is “what is the worst thing that happens if it does this wrong at three in the morning?”
Delegate the work. Never delegate the accountability.
This is not legal advice. If you handle health, financial or children's data, speak to someone qualified before you start.
Run it beside yourself before you run it instead of yourself.
Do the job the way you always do. Have it do the same job in parallel. Every time it is wrong, add a line to the brief instead of correcting the output.
Keep the part it does well; take the rest back. Almost nobody's first attempt survives at full scope — that is the mechanism working, not failing.
It produces, you review. You keep the first step — choosing the material — and the last — approving what goes out.
Saving real time, honestly measured? Write it down properly and pick a second job. If not, drop it without embarrassment and try a different one.
Any of the major assistants on a paid business tier is the right first step — ChatGPT, Claude, Gemini or Grok — at roughly R400–R500 a month. Switch off training on your data in the settings. Pick one; do not shop around for six weeks. The difference between them is far smaller than the difference between using one and not.
The bigger cost is your afternoon: writing the brief, watching the first runs, fixing the instructions. Budget one afternoon per workflow and you will be about right.
Rather have one set up and managed for you, living on your WhatsApp? That is what our hosted assistant does.
Questions from the talk, or a first workflow you are not sure how to write down — email me. I read them all.
Riaan van Straaten · StratCon (Pty) Ltd · Bloemfontein · +27 72 549 1945